I'm interested to know if there is a method for comparing the performance of two different models, such as a linear model and a linear mixed effect model on the same data set? I'm interested to compare these two statistical models and comment on their suitability for interpreting the data. Any suggestions would be greatly appreciated!
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
2
Test whether the random effects are significant using the likelihood ratio test. The parameters you're testing are on the boundary of the parameter space but the LRT is actually conservative in this case so you're good. If the random effects are not significant then use the regular linear model.
-
$\begingroup$ To clarify, do you mean a LRT between the two models or is there a way to test just the random effects in the mixed model. I have been using the lmer package in R with these two models:
mixedmodel <- lme(Mass ~ Temperature + Time + Temperature* Time, random = ~1 + Time | Tank, data = input)linearmodel <- lm(Mass ~ Temperature + Time + Temperature* Time, data = input)lrtest(mixedmodel, linearmodel)$\endgroup$ Commented Feb 3, 2017 at 1:36 -
$\begingroup$ @JaredTromp you'll want to test the random slopes Time|Tank and random intercepts 1|Tank. Opinions vary but for simple cases there are cases to be made to always leave relevant grouping factors in, if for no other reason than to prevent a reviewer from bugging you about it. Finally, I forget if lm() output plays nice with lme() you nay need to use gls() $\endgroup$ Commented May 11 at 19:02